Drug development is a sequence of narrowing gates. Thousands of compounds are screened in discovery, a small fraction are selected for further study, fewer still enter preclinical testing in animals and cells, and only a handful are approved for testing in humans. At each gate, most of what entered is eliminated.
The important pattern is not simply that many candidates fail, but where they fail. A large share of attrition happens in the early stages, during discovery and preclinical work, where failure is relatively cheap. A second, more painful band of failure happens later, during the human trial phases, where each remaining candidate has already absorbed years of work and substantial spending. The funnel therefore has two distinct regions: an early region where failure is expected and affordable, and a late region where failure is rare in number but severe in consequence.
Reading the funnel this way reframes the whole problem. The question is not whether candidates will fail, because most will. The question is whether we can move the detection of doomed candidates earlier, into the cheap region of the funnel, before they consume the resources of the expensive region.